The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to execute shell scripts (.sh) located in temporary directories ($TMPDIR) during the handling of compiler crashes. While intended for reproducing crashes, executing scripts from computed paths is a significant risk if the tool output or environment is compromised.
[DATA_EXFILTRATION]: The instructions require the agent to access sensitive system log directories, specifically ~/Library/Logs/DiagnosticReports/, to retrieve compiler crash reports. Accessing files within a user's home directory can lead to the unauthorized exposure of system state and diagnostic information.
[COMMAND_EXECUTION]: The skill performs automated version control operations (detecting VCS, running status, diff, stage, and commit commands) and build system interactions (executing xcodebuild or make). These operations represent high-privilege actions that modify the local filesystem and project state.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the user's codebase, including C source files and build configurations (.xcodeproj, project.pbxproj, Makefile), which could influence agent behavior.
Ingestion points: User-provided source code and project metadata read via the Read and VCS tools.
Boundary markers: No explicit delimiters or boundary instructions are used to separate user-provided data from agent instructions.
Capability inventory: The agent has the capability to execute arbitrary shell commands via VCS/build tools and to run shell scripts via sub-agents.
Sanitization: The skill lacks explicit sanitization or validation of the ingested code or the contents of crash reproducer scripts before they are handled.

adopt-c-bounds-safety